Design of an iterative method for enhanced fruit detection in smart farming using YoLoV9s with transfer learning analysis

Abstract

Being able to accurately detect and classify fruits on trees would be important for smart farming in terms of the efficiency of yield management and labour cost savings. Conventional methods are prone to inconsistency in environmental conditions, resulting in lower precision in detection accuracy. This paper proposes a state-of-the-art fruit detection technique through an approach to image augmentation enhanced by a genetic algorithm and advanced deep learning models. The proposed approach works for the detection of chickoo, mango, sweet lime and tomato fruit. This approach employs a genetic algorithm to optimize image augmentation, which enhances the diversification of training data and also increases the adaptability of the model towards various conditions. Fruit detection will be performed using the newest You Only Look Once (YoLoV9s) framework, which provides the possibility of real-time detection of objects. Classification of the fruits will be done with the help of a transfer learning-based VGGNet16 model. A colour thresholding step is applied for further confirmation of fruit types; it gives the least number of wrong classifications. It can be shown from the experimental results that the model that I have proposed accords a very good level of accuracy (97.9%), precision (97.0%), recall (97.5%), and an F1-score of 97.3%, which is much beyond the different methodologies produced in the literature up till now. The model's real capability, that is, its ability to accurately predict fruits under various environmental conditions reflects its potential to increase productivity in the precision agriculture process.